[WIP] Add datasets to config api V2

[leoromanovich]

• I've checked contribution guide.

[AlekseySh]

I like this approach. Let's continue with this.

As the next step, I think you need to update pipelines code with usage of a new split argument

[AlekseySh]

let's use "df" name here? so it's more consistent with argument names, but replace it with read actual df in runtime

[AlekseySh]

I suggest to rework this function a bit:

def get_dataset_by_cfg(cfg: TCfg, split: Optional[str] = None) -> IBaseDataset:
    if cfg['name'] in DATASETS_REGISTRY:
        df = pd.read_csv(Path(cfg["args"]["dataset_root"]) / cfg["args"]["df"], index_col=False)

        mapper = {l: i for i, l in enumerate(df.sort_values(by=[SPLIT_COLUMN])[LABELS_COLUMN].unique())}
        df[LABELS_COLUMN] = df[LABELS_COLUMN].map(mapper)

        if split is not None:
            df = df[df[SPLIT_COLUMN] == split].reset_index(drop=True)

        cfg["args"]["df"] = df

    else:
        if split is not None:
            raise ValueError("We only support <split> option for built-in datasets.")

    if split and "dataframe_name" in cfg["args"].keys():

    return dataset_class(**cfg["args"])

I also removed check_retrieval_dataframe_format because there was no such functionality before
I also hope in my implementation we dont need filtering arguments